•Porous materials discovery with machine learning and genetic algorithm.•Manufacturing recipe explored to improve permeability and filtration efficiency.•Evolutionary history-based use of reduced design space for material performance. This study proposes a material discovery framework for porous materials to identify design variable recipes and the corresponding material structures that can be utilized to improve the actual manufacturing process. The effectiveness of the proposed framework has been demonstrated via multi-objective genetic algorithm optimization with regard to permeability and filtration efficiency. A simulation model to generate porous material structures with two layers has been developed with design variables, such as grain diameter, grain shape, and ratio of pore former to base grain. The design variables have been optimized to maximize two objective functions, that is, permeability and filtration efficiency, which have been evaluated by machine-learning-based surrogate models with negligible computational cost as compared to the computational fluid dynamics (CFD) simulations. The surrogate models are updated once or regularly using the generated structures to improve the exploration capability according to the necessity of the optimization process. The proposed framework successfully unveiled design variable recipes and guidelines for obtaining preferable structures with high permeability and filtration efficiency in the actual manufacturing process.